bypass-finals icon indicating copy to clipboard operation
bypass-finals copied to clipboard

Using Infection library fails because of another stream wrapper

Open romm opened this issue 5 years ago • 12 comments

The library Infection uses its own stream wrapper, see \Infection\StreamWrapper\IncludeInterceptor.

Because of that, the mutants created by the library are not used and that makes the whole library unusable if the tests use bypass-finals, because the stream wrapper is overriden.

I tried to guess how both stream wrappers could be used alongside each other but as I have no expertise at all in this topic it got hard very fast.

Do you think this could be possible? How could it be done? Can I help 🙂 ?

romm avatar Feb 12 '19 16:02 romm

Maybe chaining like BypassFinals::enable($inner = IncludeInterceptor::class).

milo avatar Feb 18 '19 18:02 milo

@milo enable doesn't take any parameter? Is this correct?

bartoszkubicki avatar Oct 11 '19 19:10 bartoszkubicki

It is theorethical propose only.

milo avatar Oct 11 '19 19:10 milo

TL;DR

Same problem here

More

Custom PHPUnit hook raises an issue with a lot of escaped mutants created by Infection. Before implementation of the hook (and without call of BypassFinals::enable();) all mutants were killed.

Details below 👇

declare(strict_types=1);

namespace App\PHPUnit\Hook;

use DG\BypassFinals;
use PHPUnit\Runner\BeforeTestHook;

final class BypassFinalsHook implements BeforeTestHook
{
    public function executeBeforeTest(string $test): void
    {
        BypassFinals::enable();
    }
}
<?xml version="1.0" encoding="UTF-8"?>

<!-- https://phpunit.readthedocs.io/en/latest/configuration.html -->
<phpunit xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
         xsi:noNamespaceSchemaLocation="vendor/phpunit/phpunit/phpunit.xsd"
         backupGlobals="false"
         colors="true"
         bootstrap="tests/bootstrap.php"
         convertDeprecationsToExceptions="false"
         executionOrder="random"
>
    <!-- ... -->

    <extensions>
        <extension class="App\PHPUnit\Hook\BypassFinalsHook"/>
    </extensions>
</phpunit>

kniziol avatar Sep 14 '21 12:09 kniziol

For reference, see

  • https://github.com/infection/infection/issues/1275#issuecomment-647138464
  • https://github.com/infection/infection/issues/1275#issuecomment-647180895
  • and not merged PR: https://github.com/infection/include-interceptor/pull/13

I didn't have a chance to finish it or came up with another idea, but this is a good start for further thinking

maks-rafalko avatar Sep 14 '21 12:09 maks-rafalko

After spending hours trying to fix it on Infection side, I think it will be easier to fix it on this library, making another class that extends BypassFinals or just patching BypassFinals itself. Yeah, I understand that from SRP and just logically - BypassFinals should not be aware of Infection, same way as Infection should not be aware of BypassFinals, but since both libs are working with Stream Wrappers and stream wrappers are very easy being overridden - I see no other solution.

The simplest one that I came up with is the following:

https://github.com/dg/bypass-finals/blob/495f5bc762e7bf30a13ed8253f44bb3a701767bb/src/BypassFinals.php#L147

update this line of code and ask Infection for already overridden file, fetching $content not from the original source code, but from already mutated code by Infection.

Pseudo code would be the following:

public function stream_open(string $path, string $mode, int $options, ?string &$openedPath): bool
{
    // ...

    if (getenv('INFECTION') === 1 && InfectionIncludeInterceptor::hasPath($path)) {
        $content = InfectionIncludeInterceptor::getOverridden($path);
    } else {
        // existing BypassFinals code ...
        $content = $this->native('file_get_contents', $path, $usePath, $this->context);
    }

    // ...
}

if @dg is interested in something like this, I can update infection/include-interceptor to support this new API methods and make PoC for BypassFinals.

maks-rafalko avatar Oct 03 '21 12:10 maks-rafalko

Hi @maks-rafalko. What do you think, would something like this work? That would just be BypassFinal run as a second one and it tries to call the previously defined wrapper.

https://github.com/dg/bypass-finals/commit/7c08b98712099224a53bdeb36a58f66e5cf73e61

dg avatar Oct 03 '21 15:10 dg

I like the idea, however it doesn't work on my machine. How I tested it:

  • created a branch with e2e test on Infection repository
  • installed https://github.com/dg/bypass-finals/commit/7c08b98712099224a53bdeb36a58f66e5cf73e61
  • ran some tests

How to reproduce and play with it:

git clone [email protected]:infection/infection.git
cd infection
git checkout bugfix/include-interceptor

# install Infection's dependencies
composer install 

# go to e2e test directory
cd tests/e2e/Stream_Wrapper_Execution

# install e2e test dependencies
composer install

XDEBUG_MODE=coverage ../../../bin/infection --debug --log-verbosity=all

after that, if you open text.log file to see what is the output of PHPUnit executed under Infection with BypassFinals enabled:

I see the following error:

PHPUnit 9.5.9 by Sebastian Bergmann and contributors.
  
  Class "PHPUnit\Util\ErrorHandler" not found

Also, when I dump $meta, there is no wrapper_data key:

array(9) {
  ["timed_out"]=>
  bool(false)
  ["blocked"]=>
  bool(true)
  ["eof"]=>
  bool(false)
  ["wrapper_type"]=>
  string(9) "plainfile"
  ["stream_type"]=>
  string(5) "STDIO"
  ["mode"]=>
  string(1) "r"
  ["unread_bytes"]=>
  int(0)
  ["seekable"]=>
  bool(true)
  ["uri"]=>
  string(112) "/path/to/infection/tests/e2e/Stream_Wrapper_Execution/vendor/dg/bypass-finals/src/BypassFinals.php"
}

It is PHP 8.0.10

maks-rafalko avatar Oct 03 '21 16:10 maks-rafalko

@maks-rafalko It looks like you can't register another wrapper inside stream_open...

I tried a different approach https://github.com/dg/bypass-finals/commit/aa882485188494ae831da14d1139ed3bdd161e41

dg avatar Oct 04 '21 11:10 dg

@dg this is much better!

I tested it for the repository/branch/e2e test I mentioned above and found 1 bug that 1 mutant from 5 still escaped. I was able to fix it and let me try to explain what happens.

TL;DR: everything works with your patch https://github.com/dg/bypass-finals/commit/aa882485188494ae831da14d1139ed3bdd161e41 combined with the patch below

In details:

in that e2e test I've created, there is a class SourceClass that is not final (it's important) and for this file Infection does the following mutation

- public function getOne(): int
+ protected function getOne(): int

however it escapes, because original file is being used instead of mutated one. Why it happens? Let's closer look to your patched BypassFinal, here

https://github.com/dg/bypass-finals/commit/aa882485188494ae831da14d1139ed3bdd161e41#diff-b0de2d102039b11e3840a14e052c230cfa1377774ce14d292adca84bcbccfbd9R169-R179

So, when self::$prevWrapper is used and not null, it loads mutated version of SourceClass thanks to these lines.

However, then it check $modified !== $content which false, because SourceClass does not have final keyword and BypassFinal doesn't created tmp file, going forward to these lines where origin $path is used and overrides all the $content of Infection's interceptor.

So my patch is quite simple:

- if ($modified !== $content) {
+ if ($modified !== $content || (self::$prevWrapper && $content !== $this->native('file_get_contents', $path, $usePath, $this->context))) {

what it does basically is it stores in a temp file mutated (replaced) content of Infection's interceptor when original file does not contain final keyword, and replaced (intercepted from Infection) file is used instead of original one, making it properly working.

Result: with you latest patch updated with the diff above - everything works great! This is a very exciting result.

What do you think about this patch? Do you think we can have it on BypassFinal source code?

By the way, we have a Discord channel, so if you think it will be quicker to discuss it online - feel free to join and ping me https://discord.gg/ZUmyHTJ

maks-rafalko avatar Oct 09 '21 20:10 maks-rafalko

ping @dg, in case you missed it

maks-rafalko avatar Nov 07 '21 14:11 maks-rafalko

@dg is there anything I can help with?

we are getting complaints from users that do not understand why Infection does not work when it should, and then it boils down to bypass-final usage, so we decided to add a conflict section with bypass-final in our composer.json: https://github.com/infection/infection/pull/1605

However, personally I don't like it, as it makes user to choose either to use infection but not this lib, or vice-versa.

At the same time, I don't want to spend time making a PR if you are not going to accept it with a solution above, so if you find a minute, could you please leave your feedback here?

Alternatively, we (Infection) can add another package that will be a wrapper around bypass-final and do there the job from above comments, so users will have to install infection/bypass-final instead of this lib. Again, it will solve the issue, but I would like to avoid it and rather complete the job here to support bypass-final and Infection together natively.

Also, if you don't have enough time, does it make sense to add more collaborators to this project?

Thank you.

maks-rafalko avatar Nov 26 '21 15:11 maks-rafalko

@maks-rafalko I've tried the patch and it doesn't seem to work for me anymore, will likely debug this further and create a PR.

dkarlovi avatar Aug 31 '22 07:08 dkarlovi

Note: after opening #33, I've noticed that master incorporates the patch made by @dg here and it actually works with Infection 0.26.13 for me. I think this can be closed as soon as a bugfix release is made.

dkarlovi avatar Sep 02 '22 10:09 dkarlovi

@dkarlovi I still reproduce the issue explained with details in my comment here: https://github.com/dg/bypass-finals/issues/9#issuecomment-939357946

So,

  • yes, I confirm that most of the needed code is already in master branch of this repo
  • no, I can't confirm that it works
  • my patch from that comment is still needed, thus your PR (https://github.com/dg/bypass-finals/pull/33) should be reopened

How you can test it yourself?

Look at this branch https://github.com/infection/infection/compare/bugfix/include-interceptor#diff-5447adae4da9395e026519944f362f9f7cd6986f5e5745677c9da1cbb4a3e672R5 and added e2e test.

git clone [email protected]:infection/infection.git
cd infection
git checkout bugfix/include-interceptor

# install Infection's dependencies
composer install 

# go to e2e test directory
cd tests/e2e/Stream_Wrapper_Execution

# install e2e test dependencies
composer install

../../../bin/infection --debug --log-verbosity=all

1 Mutant is escaped (which is wrong). If you manually apply the patch in /tests/e2e/Stream_Wrapper_Execution/vendor/dg/bypass-finals/src/BypassFinals.php:

- if ($modified !== $content) {
+ if ($modified !== $content || (self::$prevWrapper && $content !== $this->native('file_get_contents', $path, $usePath, $this->context))) {

then it works as expected - all the mutants are killed.

What are the next steps?

  • let's reopen https://github.com/dg/bypass-finals/pull/33 and get it merged
  • then do a release of this package
  • then we will update conflicts section in infection/infection to allow new versions of dg/bypass-finals

https://github.com/infection/infection/blob/d35e0795d64c2a1e030e63c51fbf6b37991b08d2/composer.json#L71

maks-rafalko avatar Sep 03 '22 10:09 maks-rafalko

PR reopened.

dkarlovi avatar Sep 05 '22 08:09 dkarlovi

Sorry I didn't have time to do this sooner, I maintain a lot of open source projects and I have hundreds of open issues, so everything takes annoyingly long time.

I tried to create a new solution and it seems that test tests/e2e/Stream_Wrapper_Execution passes.

dg avatar Sep 06 '22 05:09 dg

@dg I can confirm this commit fixes the issue and e2e test above is green!

could you make a release please? The last one was on September 2020. Then we can remove conflict for dg/bypass-finals on infection/infection side.

Sorry I didn't have time to do this sooner, I maintain a lot of open source projects and I have hundreds of open issues, so everything takes annoyingly long time.

No need to apologize. The only one recommendation - does it make sense to add more collaborators you trust? So other developers can help with reviewing/merging/releasing.

Thanks you.

maks-rafalko avatar Sep 10 '22 11:09 maks-rafalko

Good to hear that it works, I've released a new version.

dg avatar Sep 13 '22 01:09 dg